The following paper aims to introduce into the pedagogical landscape a critique of the function of LLMs, large language models, as mediators in relational and teaching practices. It will explore the genesis and functionality of Generative Artificial Intelligence and, in particular, Large Language Models (LLMs), which are now widely used in formal, non-formal, and informal education settings. Following a brief theoretical introduction to the devices, two illustrative cases are presented: Humy.AI and Replika. These two companies propose the use of a ‘relational agent’ which, starting from a specific LLM, performs similar tasks but in very different fields: Humy.AI is developed to implement history teaching, while Replika is a relational agent, or AI companion, developed to function as a virtual companion, capable, for example, of empathic listening and non-judgmental dialogue. The criticism is based on the fact that, in both cases, a large language model, simply because it can imitate natural language and respond effectively from a syntactic point of view, becomes an epistemological actor and decision-maker in the growth processes of individuals in training, just as it is when it comes to or-ganizing certain processes quantitatively. For this reason, it is urgent to understand to what extent – and to what degree – an agent based on generative artificial intelligence can be a mediator in these processes. Where empirical research must clearly identify the terms of this revolution, it is equally important to take a clear po-sition and not confuse the pedagogical tasks that every educator must respond to, in order to avoid excessive delegation to algorithms and agents based on generative artificial intelligence, as is already the case in certain specific areas of human resources, robotics, and algorithmic workplace management.
Il seguente articolo si pone l’obbiettivo di introdurre nel panorama pedagogico una critica alla funzione dei modelli di linguaggio come mediatori nelle pratiche relazionali e in quelle didattiche. Verrà esplorata la ge-nesi e la funzionalità dell’intelligenza artificiale generativa e in particolare degli LLMs, Large Language Mo-dels, oggi ampiamente utilizzati sia nei luoghi dell’educazione formale che di quella non formale e informale. A seguito di una breve introduzione teorica dei dispositivi, vengono proposti due casi esemplificativi: Humy.AI e Replika. Queste due aziende propongono di utilizzare un “agente relazionale” che, a partire da uno specifico LLM, svolge compiti simili ma in ambiti molto diversi: Humy.AI è sviluppata per implementare la didattica della storia, mentre Replika è un agente relazionale, o AI companion, sviluppato per funzionare come un compagno o compagna virtuale, capace, ad esempio, di ascolto empatico e dialogo non giudicante. La critica si articola sul fatto che, in entrambi i casi, un modello di linguaggio, per il semplice fatto di saper imitare il linguaggio naturale e rispondere efficacemente da un punto di vista sintattico, diventa attore epi-stemologico e decisore nei processi di crescita di individui in formazione, esattamente come lo è dove si tratta di organizzare quantitativamente determinati processi. Per questo motivo risulta urgente comprendere fin dove – e in che misura – un agente basato su intelligenza artificiale generativa possa essere mediatore in questi processi. Laddove le ricerche empiriche devono individuare con chiarezza i termini di questa rivolu-zione, risulta parimenti importante prendere una posizione chiara e non confondere i compiti pedagogici a cui ogni educatore deve rispondere, al fine di evitare un’eccessiva delega ad algoritmi e agenti basati su intelligenza artificiale generativa, come già accade negli ambiti di alcune pratiche specifiche delle risorse umane, nella robotica e nella gestione algoritmica dei luoghi di lavoro.
Astorri, G. (2025). The polysemy of generative artificial intelligence = La polisemia dell’intelligenza artificiale generativa. FORMAZIONE & INSEGNAMENTO, 23(S1 (numero speciale 1)), 60-66 [10.7346/-feis-XXIII-01-25_10].
The polysemy of generative artificial intelligence = La polisemia dell’intelligenza artificiale generativa
Giacomo Astorri
2025
Abstract
The following paper aims to introduce into the pedagogical landscape a critique of the function of LLMs, large language models, as mediators in relational and teaching practices. It will explore the genesis and functionality of Generative Artificial Intelligence and, in particular, Large Language Models (LLMs), which are now widely used in formal, non-formal, and informal education settings. Following a brief theoretical introduction to the devices, two illustrative cases are presented: Humy.AI and Replika. These two companies propose the use of a ‘relational agent’ which, starting from a specific LLM, performs similar tasks but in very different fields: Humy.AI is developed to implement history teaching, while Replika is a relational agent, or AI companion, developed to function as a virtual companion, capable, for example, of empathic listening and non-judgmental dialogue. The criticism is based on the fact that, in both cases, a large language model, simply because it can imitate natural language and respond effectively from a syntactic point of view, becomes an epistemological actor and decision-maker in the growth processes of individuals in training, just as it is when it comes to or-ganizing certain processes quantitatively. For this reason, it is urgent to understand to what extent – and to what degree – an agent based on generative artificial intelligence can be a mediator in these processes. Where empirical research must clearly identify the terms of this revolution, it is equally important to take a clear po-sition and not confuse the pedagogical tasks that every educator must respond to, in order to avoid excessive delegation to algorithms and agents based on generative artificial intelligence, as is already the case in certain specific areas of human resources, robotics, and algorithmic workplace management.| File | Dimensione | Formato | |
|---|---|---|---|
|
FI_2025_23_S1_60-66_Astorri.pdf
accesso aperto
Tipo:
Versione (PDF) editoriale / Version Of Record
Licenza:
Licenza per Accesso Aperto. Creative Commons Attribuzione (CCBY)
Dimensione
977.14 kB
Formato
Adobe PDF
|
977.14 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.



